A 108nW 0.8mm2 Analog Voice Activity Detector (VAD) Featuring a Time-Domain CNN as a Programmable Feature Extractor and a Sparsity-Aware Computational Scheme in 28nm CMOS

📋 论文概要

该论文提出了一种108nW、0.8mm2的模拟语音活动检测器（VAD），采用时域卷积神经网络（CNN）作为可编程特征提取器，解决了传统VAD因全带宽高分辨率数据转换导致功耗过高的问题，实现了超低功耗始终在线语音检测。

💡 主要创新点

• 提出模拟域时域CNN特征提取器，避免数字域高功耗数据转换，实现极低功耗语音活动检测。
• 采用可编程架构，使特征提取器能灵活适应不同语音环境，提升检测精度。
• 整体VAD仅消耗108nW功率，面积仅0.8mm2，适合可穿戴设备中的始终在线应用。

🏷 关键词

📄 原文摘要

University of Lisboa, Lisbon, Portugal 1 2 An ultra-low-power always-on voice activity detector (VAD) is the key enabler of acoustic sensing in wearables. The VAD listens to the environment and wakes up the main system only when there is a right activity detected. Since most human-centric applications have infrequent activities, the VAD dominates the system power. The traditional VAD using the digital feature extractor and classifier [1] requires full-bandwidth and high-resolution data conversion before digital-signal processing, drawing a substantial power (>20µW). Recently, the analog feature extractor shows more promises in power reduction. In [2, 3], the analog-filter bank brings the feature-extraction power down to 1µW (Fig. 22.5.1, upper). Yet, the analog-filter bank does not support reprogramming and has a large area (~0.1mm2/channel) that limits the number of input channels of the following deep neural

👥 作者与机构

Feifei Chen1, Ka-Fai Un1, Wei-Han Yu1, Pui-In Mak1, Rui P. Martins1,2

University of Macau, Macau, China